# Research Method Policy

## Evidence Policy

Model output, hypotheses, search snippets, and drafts are not evidence. Verified evidence uses stable IDs and recorded provenance.

## Literature Search Policy

Record exact sources, queries, filters, access dates, counts, retained identifiers, failures, and coverage limits.

## Research Position Policy

Do not create a substantive study protocol until a bounded search, source-level evidence map, nearest-work comparison, contribution statement, feasibility review, and adversarial review support an `advance` or `pilot-only` decision.

## Claim Lifecycle

Claims move through candidate, planned, supported, mixed, unsupported, or withdrawn states without deleting contradictory history.

## Protocol Revision Policy

Frozen protocols are never rewritten in place. Amendments and revisions record timing, owner, rationale, and prior result access.

## Iteration Policy

Refuting, null, mixed, failed, and inconclusive outcomes remain visible. Follow-up hypotheses derived after results are exploratory until tested using new target data under a new frozen protocol. Runs are never repeated merely to obtain a favorable result.

## Analysis Classes

Every analysis is labeled confirmatory, exploratory, or post-hoc.

## Reproducibility Policy

Runs and analyses retain code, data, environment, configuration, seed, command, output, and deviation provenance.

## Confidentiality and External Services

unresolved

## Human Approval Gates

unresolved